AI derived ECG global longitudinal strain compared to echocardiographic measurements

Area of Science:

• Cardiology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Left ventricular global longitudinal strain (LVGLS) is a crucial metric for assessing cardiac function but is challenging to obtain.
• Heart failure (HF) diagnosis and prognosis rely on accurate assessment of LV systolic function.
• Novel, accessible methods for evaluating LV systolic function are needed.

Purpose of the Study:

• To evaluate an AI-generated electrocardiography score for LVGLS estimation (ECG-GLS score).
• To assess the utility of the ECG-GLS score in diagnosing LV systolic dysfunction.
• To determine the prognostic value of the ECG-GLS score in patients with heart failure.

Main Methods:

• A deep-learning algorithm (convolutional neural network) was developed to estimate LVGLS from ECG data.
• The performance of the ECG-GLS score was validated using data from an acute HF registry (n=1186).
• The ECG-GLS score's ability to identify impaired LVGLS (≤12%) and low LVEF (<40%) was assessed using ROC analysis.

Main Results:

• The ECG-GLS score effectively identified patients with impaired LVGLS (AUROC=0.82) and was comparable to LVGLS in identifying patients with LVEF <40% (AUROC=0.85 vs 0.83).
• Low ECG-GLS scores were significantly associated with increased 5-year all-cause mortality and HF hospitalizations.
• The ECG-GLS score proved to be a significant independent risk factor for adverse outcomes in HF patients.

Conclusions:

• The AI-generated ECG-GLS score is a practical and effective tool for estimating LVGLS.
• This score demonstrates significant potential for diagnosing LV systolic dysfunction and predicting long-term prognosis in heart failure patients.
• The ECG-GLS score may serve as a valuable, accessible alternative to traditional LVGLS measurements in clinical practice.